---
title: 'Gemini 4 Hydrogen + Antigravity is WILD'
source: 'https://youtube.com/watch?v=XUUbeYlCrKA'
video_id: 'XUUbeYlCrKA'
date: 2026-11-17
duration_sec: 665
channel: 'Julian Goldie SEO'
---

# Gemini 4 Hydrogen + Antigravity is WILD

> Source: [Gemini 4 Hydrogen + Antigravity is WILD](https://youtube.com/watch?v=XUUbeYlCrKA)

## Summary

The video demonstrates how to use Google's Gemini 4 Hydrogen model within the Antigravity platform to build three business tools, each visualized as a theme park ride. The creator shows how to generate a lead prospecting coaster, a content planning ferris wheel, and an automation monitoring tower, all from single prompts. The video also provides setup instructions and guidance on when to use Hydrogen versus the orchestrator model.

### Key Points

- **Theme Park Concept** [00:02] — The video introduces a theme park built with Google Antigravity, where each ride represents a business tool: a roller coaster for lead prospecting, a ferris wheel for content planning, and a drop tower for automation monitoring.
- **Gemini 4 Hydrogen Release** [00:28] — Google released Gemini 4 Hydrogen on November 12th, described as a 'daily driver' for agentic workflows. It's the fast model in the Gemini 4 family, finishing tasks 2.5 times faster than Gemini 3.8 Flash.
- **Performance Benchmarks** [00:43] — On Terminal Bench 4.1, Hydrogen scores 57% compared to 3.8 Flash's 18%. On the Artificial Analysis Intelligence Index, it scores 51, one point above Gemini 4 Argon.
- **Accessing Hydrogen in Antigravity** [01:13] — Hydrogen is available in Antigravity's model selector under the prompt box. It offers four thinking levels: low, medium, high, and extra. The creator chose medium for speed and set the agent to turbo mode to avoid confirmation prompts.
- **Lead Coaster Setup** [02:08] — The lead coaster is designed for agencies and automation builders. The creator placed 10 prospect notes in a folder, each with short observations about a business, and then prompted Hydrogen to generate fit scores, problems to fix, and personalized openers.
- **Lead Coaster Build Process** [03:18] — Hydrogen read the notes, wrote a leads.json file, and created the coaster visualization. The first version took about 2 minutes, and subsequent passes improved the visual style while preserving data.
- **Lead Coaster Output** [04:04] — Each lead is represented as a coaster car with a robot passenger. The ride includes stations for research, fit score, opener, and ready to send, with glass cards showing real data. Example: Northstar Coaching scored 95 due to frequent LinkedIn posts but lacking a booking page.
- **Content Ferris Wheel Setup** [04:44] — The ferris wheel is for coaches and course creators. Given a brief about a Notion course for freelancers, Hydrogen generated 12 pieces of content (3 per week for 4 weeks) across community, email, and LinkedIn, with varied formats.
- **Ferris Wheel Output** [05:30] — The wheel displays each post's day, channel, and hook line. Examples include 'The three column notion view that stops client work from slipping' and a Friday check-in prompt. The build took about 2.5 minutes and worked first time.
- **Proof Tower Problem** [06:12] — The proof tower addresses the issue that automations can report success but produce no useful output. A member noted, 'Useful output is not the same as finished work.'
- **Proof Tower Script** [06:42] — Hydrogen wrote a Python script to check the newest file in each of six automation folders, marking them as pass, late, empty, or missing. It then generated a morning note highlighting only items needing attention.
- **Proof Tower Results** [07:14] — Four of six automations passed. The nightly backup's newest file was 88 hours old, and the welcome emails log was empty (zero bytes). The tower visualization uses glowing rings with mint for pass and coral for issues, and a drop animation for failures.
- **When to Use Hydrogen vs Orchestrator** [07:55] — Hydrogen is ideal for quick scripts, small tools, tests, and docs updates. For larger projects, use the orchestrator (Carbon or Argon) which plans first, then Hydrogen executes. This combination achieves near-Carbon success rates with 24% less quota and 19% faster completion.
- **Step-by-Step Setup** [09:06] — Steps: 1) Create a new project in Antigravity with an empty folder. 2) Drop input data (prospect notes, briefs, etc.) into the folder. 3) Select Gemini 4 Hydrogen (medium is a good start). 4) Set agent permissions to turbo. 5) Paste prompt asking for data file first, then visual, then screenshot and fix overlaps. 6) Open the generated file in browser.
- **Which Ride to Build First** [09:49] — Agencies should start with the lead coaster; coaches with the ferris wheel; those managing client automations with the proof tower. For online sellers, the coaster works for wholesale outreach and the ferris wheel for product launches.
- **Mistakes to Avoid** [10:27] — Don't skip the data step (results in made-up content), keep each prompt to one ride, and always review the output before sending to real people. The opener is a first draft.

### Conclusion

Gemini 4 Hydrogen in Antigravity enables rapid creation of functional business tools from simple prompts, with visualizations that make data engaging. The key is to provide real data upfront and use the appropriate model for the task's complexity.

## Transcript

Google Anti-gravity, and I used it to build a whole theme park. Every ride in this park is [music] a real business tool. A roller coaster that turns a prospect list into personal openers, a ferris wheel that writes a month of
content for a [music] course launch, and a drop tower that checks your automations actually finish their work and tells you when one failed [music] quietly. Hydrogen built every one of them on camera from one prompt each.
made, and how you can build the same rides for your business today. Here's what happened. On November the 12th, Google released Gemini 4 Hydrogen. Google calls it their daily driver for Aentic Workflows. It's the fast model in
the Gemini 4 family. The numbers from Google's announcement. It finishes tasks 2 and a half times faster than Gemini 3.8 Flash. On Terminal Bench 4.1, it 3.8 Flash. On Terminal Bench 4.1, it scores 57% where 3.8 8 Flash scored
about 18. And on the artificial analysis intelligence index, it scores 51, one point above Gemini 4 Argon. So, it's quick and it's smart enough to build real things on its own. And it's inside anti-gravity right now. You open the
model selector under the prompt box and there it is. Gemini 4 hydrogen. Hover over it and you get four thinking levels. Low, medium, high, and a new one called extra. I picked hydrogen on medium because I wanted speed. Switched
the agent to turbo so it doesn't stop to ask before every step and made a new project called park. Why build a theme park? Because this week in the AI profit boardroom, the same three problems kept coming up. People asking for the best
prospecting workflow. People who built something great but can't get leads. And one member who shared a small win. He now gets told when last night's backup didn't land. because a backup can fail quietly and you only find out when you
need it. So, each ride solves one of those. Let's ride the first one. Ride number one, the lead [clears throat] coaster. This one is for agencies and AI automation builders who need more conversations with the right people. I
put 10 prospect notes in a folder. A coach with no booking page. A course owner filling a new cohort. A software founder hiring a first marketer. A Shopify store with thin product pages. Short notes, the kind you jot down after
looking at someone's site. Then I typed one prompt. Read every note. For each prospect, write a fit score with a reason. The one problem we can fix and a short personal opener that mentions a real detail from their note. then turn
it all into a 3D roller coaster at golden hour. If you want my best setups for anti-gravity, like this lead coaster prompt, the prospect note template, and the follow-up prompts I use to polish every ride, they're all inside the AI
profit boardroom. Inside, you also get the full anti-gravity course, the faceless video skill that made this video, and weekly coaching calls where you can bring your own build and get help with it. There's a big community of
agency owners, coaches, course creators, and automation builders sharing what's working for them right now. The link is in the comments and the description, or go to aiprofitboardroom.com. Right, let's watch it build. Watch the
agent work. It reads the 10 notes, writes leads.json, then writes the whole coaster, and opens it in the browser to check it. The first version took about 2 minutes. It worked, but it looked a bit flat. So, I asked for a premium golden
hour look, a warmer sky, candy glass rails, a ferris wheel and tents around the track, and bigger cards. Then, one more pass for richer color. Each pass took a minute or two, and it kept all the data exactly as it was. And here's
the ride. Every lead is a coaster car with a little robot passenger wearing a name tag. The car climbs through four stations. Research, fit score, opener, ready to send. At each station, a glass card pops up with the real data. Here's
the fit score and why. Northstar Coaching scored 95 because they post on LinkedIn three times a week, but every inquiry lands in a plain contact form instead of a booking page. And here's the opener it wrote at the last station.
It mentions the exact detail from the note, so it doesn't read like a template. The car hits ready to send and you get a little burst of confetti. The counter in the corner counts up as each car finishes. 10 leads boarded. An
car finishes. 10 leads boarded. An average fit of 87. 10 openers written. That's your outreach list for the day ranked with a first line ready to go. Ride number two, the content ferris wheel. This one is for coaches,
consultants, and anyone running a course or a community. I gave Hydrogen a short brief, an online course that teaches freelancers to run their business in Notion, a school community with weekly calls. The goal is to fill a new cohort
that starts in 4 weeks. My prompt asked for 12 pieces of content, three a week for 4 weeks across the community, email, and LinkedIn. Each one different. A story, a quick tip, a member question, a behind the scenes, a checklist. Then the
ride, a giant glass ferris wheel with 12 gondilas. As the wheel turns, the gondola at the top opens into a card with that post's day, channel, and hookline. And along the bottom, a calendar strip fills in as each post
lands. Look at the hooks it wrote. The three column notion view that stops client work from slipping. A Friday check-in that asks which notion page you open first every morning. They sound like the brief. short sentences,
friendly, practical, and it mixed up the formats like I asked, so the month doesn't feel like the same post 12 times. This one worked first time. About 2 and 1/2 minutes from prompt to a spinning wheel. This is the part members
keep asking about, not more ideas. A month that's already planned, so you can just show up and post. Ride number three, the proof tower. This is for anyone running automations for themselves or for clients. Here's the
problem. An automation can say it ran and still produce nothing. One member put it perfectly this week. Useful output is not the same as finished work. So, I made a folder with six automations, a nightly backup, a client
SEO report, a lead enrichment sweep, member welcome emails, a weekly content calendar, and support inbox triage. Each one is supposed to drop a file into its own folder on a schedule. My prompt asked Hydrogen to write a small Python
script that checks the newest file in each folder. Pass if a real file landed on time, late if it's too old, empty if the file is zero bytes, missing if nothing is there. Then save a short morning note with only what needs
attention. Watch it write check. py and run it. Four of the six passed. Two need attention. The nightly backup's newest file is 88 hours old when it should land every 24. And the welcome emails log landed, but it's empty. Zero bytes. So
that run said it finished and did nothing. Then the tower, one glowing ring per automation. A little inspector robot with a clipboard rides the lift up and checks each ring. Mint means pass.
Coral means something's wrong. When it hits a failed ring, the car does a gentle drop. And at the end, the morning notes slides in with just the two things you need to fix. The first version was a bit thin and far away, so I asked
Hydrogen to make the tower chunky. Bring the camera close and put the rest of the park around it. That's the members win from this week as a ride you can actually look at every morning. Now, a quick word on when to use hydrogen and
when to use something else because anti-gravity gives you a few choices in that menu. Hydrogen on its own is great for jobs like these. Quick scripts, small tools, tests, docs updates, upgrades. You want it done fast and you
can describe it clearly in one prompt. For bigger builds, there's the orchestrator. You'll see orchestrator Carbon and Orchestrator Argon in the same menu. The planner model, Carbon or Argon, explores first in readonly mode
and writes the plan. Then Hydrogen does the editing, runs the builds and tests, and checks the result. If Hydrogen gets stuck, it hands control back to the planner, which fixes the plan. Google says that combination gets close to
carbon success rate while using 24% less quotota and finishing 19% faster. And their docs recommend the orchestrator as the default for most day-to-day development. So, my rule is simple. One clear job, use hydrogen. a whole project
with lots of moving parts. Use the orchestrator. Here's how to set this up yourself step by step. One, open anti-gravity and create a new project. Point it at an empty folder. Two, drop
your inputs in that folder first. Your prospect notes, your content brief, or your automations list. The agent does much better when the real data is already there. Three, click the model selector under the prompt box and pick
Gemini 4 hydrogen. Medium is a good start. Go to high or extra if the job needs more thinking. Four, set agent permissions to turbo if you're happy for it to run commands without asking each time. Five, paste the prompt. Ask for
the data file first, then the visual, then ask it to screenshot the result and fix anything that overlaps. That last line makes a big difference. Six, open the file it made in your browser and use it. Which ride should you build first?
If you run an agency or sell automations, start with the lead coaster. Swap in your own prospect notes and you'll have openers by lunch. If you're a coach, a consultant, or you run a course or a community, start with the
ferris wheel. Give it your next launch and your voice. If you look after automations for clients, build the proof tower first. It's the one that saves you from the awkward call where the client finds the problem before you do. And if
you sell online, the coaster works for wholesale outreach and the ferris wheel works for a product launch. A few mistakes to avoid. Don't skip the data step. If you ask for the visual first, you get a pretty ride with madeup
content. Keep each prompt to one ride. And always read what it wrote before you send anything to a real person. The opener is a first draft. You're still the one who hits send. That's Hydrogen Park. Three rides, three prompts built
fast enough that you can try all three this afternoon. If you want the exact prompts for all three rides, the sample folders, and a short setup guide for each, they're inside the AI profit boardroom along with the full
anti-gravity course and the weekly calls. The link is in the comments and the description, or go to apiprofitboardroom.com. Thanks for watching, and I'll see you in the next one.
